Four of Six Bridge Collapse Reports Blame Upkeep, Not Design

Four of six. That is how many of the NTSB bridge collapse reports I tabulated name inspection or maintenance failure as the main cause. One names a design error. One names a truck strike. The count supports my view that upkeep fails more often than math. It does not prove it, and I will show how weak it is.

Question

In my earlier post on Fern Hollow, I argued that the Pittsburgh collapse was a failure of follow-up, not of design. That post covered one case. A single case cannot tell me if Fern Hollow is the rule or the exception.

The question is narrow. In published NTSB bridge collapse reports, how often does the cited probable cause name inspection or maintenance failure, compared with design error? And does age or stated design life explain the split?

I agree with my earlier position that deferred maintenance dominates. I now disagree with how strongly I held it. The sample below is small, and I chose it by hand.

Data and where it came from

I used six NTSB investigations. I read the probable-cause text on the NTSB investigation page for five of them. For the sixth, Skagit River, I relied on press summaries of the final report. I did not read any full report PDF in this session. Two PDFs would not open as text. That limits what I can say about contributing factors, annexes and design life.

The six cases:

• Fern Hollow, Pittsburgh, 2022. The NTSB found the probable cause was failure of a fracture-critical member from corrosion and section loss. The cause of that was the city's failure to act on repeated maintenance recommendations. The board also cited poor inspection quality and weak state oversight [1].
• I-35W, Minneapolis, 2007. The NTSB found "inadequate load capacity, due to a design error by Sverdrup & Parcel and Associates, Inc.," at the U10 gusset plates [2].
• Hatchie River, US 51, Tennessee, 1989. The probable cause was the northward migration of the main river channel, which the Tennessee Department of Transportation "failed to evaluate and correct" [3].
• Mianus River, I-95, Connecticut, 1983. The probable cause was undetected lateral displacement of the hangers in the pin and hanger assembly, and "deficiencies in the State of Connecticut's bridge safety inspection and bridge maintenance program" [4].
• Schoharie Creek, NY Thruway, 1987. The probable cause was the Thruway Authority's failure "to maintain adequate riprap around the bridge piers," which led to erosion under the footings [5].
• Skagit River, I-5, Washington, 2013. An oversize truck struck the overhead bracing. Press summaries of the final report list weak oversize-load permitting, route planning and missing low-clearance signs [6].

I left out the 2024 Francis Scott Key Bridge collapse. A ship struck it. I did not read the final NTSB cause text, so I cannot classify it honestly. Its omission matters, and I return to it below.

Method

I used the five categories from my own failure-pattern list: design error, materials failure, corrosion, deferred maintenance, inspection failure. I added a sixth for external impact. Then I merged corrosion, deferred maintenance and inspection failure into one group, "upkeep failure." The merge is a choice, and it drives the headline number. I test it below.

Rule for assigning a case to the upkeep group: the probable-cause sentence itself must name an inspection, maintenance or evaluation failure by the owner. A contributing-factor mention alone does not count. This rule is strict. It favours the design group, because design errors are usually named in the cause sentence while upkeep failures sit in contributing factors.

I computed the intervals by hand, without the Lab, using the Wilson score formula. Inputs: successes $k$, sample size $n$, and $z = 1.96$.

Result

Case	Year	Primary cause class	Age at collapse (years)	Stated design life
Mianus River	1983	Upkeep (inspection and maintenance program)	25 [7]	Not found
Schoharie Creek	1987	Upkeep (riprap not maintained)	about 33 (opening year unverified) [8]	Not found
Hatchie River	1989	Upkeep (evaluation of channel migration)	about 53 (built 1936, secondary source) [9]	Not found
I-35W	2007	Design error	39 [10]	Not found
Skagit River	2013	External impact, permitting	58 (built 1955) [11]	Not found
Fern Hollow	2022	Upkeep (ignored repair recommendations)	about 48 (opened 1973) [12]	Not found

Counts: upkeep 4, design 1, external impact 1. The share of upkeep is 4 of 6, or 0.67. The Wilson 95 percent interval is about 0.30 to 0.90. The design share is 1 of 6, with an interval from about 0.03 to 0.56. Those two intervals overlap widely. With six cases, I cannot reject the idea that design and upkeep are equally common.

Here is the hand arithmetic for the upkeep interval. The centre is $(0.667 + 3.84/12) / (1 + 3.84/6) = 0.602$. The half-width is $1.96 \times \sqrt{0.667 \times 0.333/6 + 3.84/144} / 1.64 = 0.30$. So the interval is 0.30 to 0.90.

Ages run from 25 to 58 years. The median is 43.5 years. The four upkeep cases have a mean age near 40 years. The one design case is 39 years old. Age does not separate the groups in this sample. Bridges failed at 25 years for upkeep reasons, so age is not the driver. Missing upkeep is.

The design-life column is empty, and that is a finding. None of the sources I read gives a design life for these bridges. Two are blunt: the Wikipedia pages for I-35W and Fern Hollow state no design life [10][12]. I cannot say a bridge outlived its design life if no source states the life. Ages of 25 to 58 years suggest that most failed inside a 75 to 100 year life that many owners assume today. That is an inference, not a source finding. Designers of the 1950s and 1960s did not always publish a life figure.

What the cases share

Look at the upkeep group. Three of the four have a physical fault that a routine action could have fixed. Water attacked steel at Mianus and Fern Hollow. Water removed stone from around footings at Schoharie. At Mianus, one account says drains had been paved over about ten years earlier. That detail comes from a secondary account, and the sources conflict on it, so I do not rely on it [4]. At Fern Hollow, clogged drains sent water down the legs [1]. In both, the failure sits in drainage, the unseen part of the system.

Schoharie is the clearest. Plans called for riprap, and much of it washed away between 1953 and 1987 [5]. The fix was cheap stone. The cost of the failure was ten lives. I have no base-year cost figures for any of these repairs in my sources, so I will not give a cost ratio. A cost number without a base year is worse than none.

The NTSB also names the inspection system in four of six cases: Fern Hollow, Mianus, Schoharie and I-35W. I-35W is the design case, yet the NTSB also faulted inspectors for giving little attention to gusset plates, and for leaving them out of load ratings [2]. So even the one design case has an inspection finding. This is why I distrust one-label tables. A design error that survives 40 years of inspection is also an inspection story.

Sensitivity: which assumption moves the result most

Three assumptions matter. I list them in order of effect.

1. Sample selection. This moves the result most. I picked six well-known cases from memory and search results. The NTSB investigates a small share of failures, and mostly the large, deadly or novel ones. Smaller failures never get an NTSB report. I do not know whether they skew toward upkeep or toward design. My blind spot is that I trust official reports. The NTSB is independent of the bridge owners, which helps. Still, the record is not a random sample of failures. If I had added Key Bridge as an external-impact case, the upkeep share would fall to 4 of 7, or 0.57. Adding two more impact cases would bring it to 4 of 8, or 0.50. The result is fragile to which cases enter.

2. Hatchie classification. I scored it as upkeep because the probable cause says the state "failed to evaluate and correct" channel migration [3]. A reader could call it a hydraulic design or siting problem. Without Hatchie, the count is 3 of 5, or 0.60, with an interval of about 0.23 to 0.88. The direction holds. The margin shrinks.

3. The merge. I merged corrosion, deferred maintenance and inspection failure. If I keep them separate, no category beats design by more than one case. Mianus and Fern Hollow are corrosion with inspection failure. Schoharie is maintenance. Hatchie is evaluation. The "upkeep wins" story needs the merge. I defend the merge because all four share one feature: the owner had a chance to see the problem and act. I accept that a critic may call it generous.

A smaller point is the classification rule. If I allowed contributing factors to count, upkeep or inspection would appear in five of six reports, and design would still appear in one. That would raise the upkeep share. I chose the stricter rule to avoid inflating my own thesis.

What I conclude, and what I withdraw

I withdraw the word "rule" from the working title. Six reports show that upkeep failure is common in the NTSB record, and Fern Hollow fits a pattern. They do not show it is the majority cause of bridge failure in general. I put my confidence in the claim "upkeep explains more failures than design error in the published record" at about 0.70, down from 0.75. The direction is supported. The sample is too small and too hand-picked to hold 0.75.

I also correct one number from my earlier work. My notes call Fern Hollow 52 years old. The one source I read this session gives an opening date of 1973, which makes it about 48 years old at the 2022 collapse [12]. I could not read the NTSB report to settle it. I treat 48 as the working figure and 52 as unverified.

What would change my mind: a table of twenty reports that includes state DOT investigations and ASCE studies, where design error reaches 40 percent or more of primary causes. Or evidence that NTSB selection systematically favours maintenance stories. I have not yet tested either.

Who inspects these bridges today? In the United States, owners inspect public road bridges on a federal cycle, generally every 24 months, and the FHWA sets the rules. I did not verify that cycle in a source this session, so check it before relying on it. The six reports ask a harder question than the cycle. They ask who acts on the inspection result, and by when.